1/* Copyright (C) 1996-2023 Free Software Foundation, Inc.
2 This file is part of the GNU C Library.
3
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published
6 by the Free Software Foundation; version 2 of the License, or
7 (at your option) any later version.
8
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
13
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, see <https://www.gnu.org/licenses/>. */
16
17#ifdef HAVE_CONFIG_H
18# include "config.h"
19#endif
20
21#include <argp.h>
22#include <assert.h>
23#include <ctype.h>
24#include <endian.h>
25#include <errno.h>
26#include <error.h>
27#include <fcntl.h>
28#include <iconv.h>
29#include <langinfo.h>
30#include <locale.h>
31#include <libintl.h>
32#include <limits.h>
33#include <nl_types.h>
34#include <obstack.h>
35#include <stdint.h>
36#include <stdio.h>
37#include <stdlib.h>
38#include <string.h>
39#include <unistd.h>
40#include <unistd_ext.h>
41#include <wchar.h>
42
43#include "version.h"
44
45#include "catgetsinfo.h"
46
47
48#define SWAPU32(w) \
49 (((w) << 24) | (((w) & 0xff00) << 8) | (((w) >> 8) & 0xff00) | ((w) >> 24))
50
51struct message_list
52{
53 int number;
54 const char *message;
55
56 const char *fname;
57 size_t line;
58 const char *symbol;
59
60 struct message_list *next;
61};
62
63
64struct set_list
65{
66 int number;
67 int deleted;
68 struct message_list *messages;
69 int last_message;
70
71 const char *fname;
72 size_t line;
73 const char *symbol;
74
75 struct set_list *next;
76};
77
78
79struct catalog
80{
81 struct set_list *all_sets;
82 struct set_list *current_set;
83 size_t total_messages;
84 wint_t quote_char;
85 int last_set;
86
87 struct obstack mem_pool;
88};
89
90
91/* If non-zero force creation of new file, not using existing one. */
92static int force_new;
93
94/* Name of output file. */
95static const char *output_name;
96
97/* Name of generated C header file. */
98static const char *header_name;
99
100/* Name and version of program. */
101static void print_version (FILE *stream, struct argp_state *state);
102void (*argp_program_version_hook) (FILE *, struct argp_state *) = print_version;
103
104#define OPT_NEW 1
105
106/* Definitions of arguments for argp functions. */
107static const struct argp_option options[] =
108{
109 { "header", 'H', N_("NAME"), 0,
110 N_("Create C header file NAME containing symbol definitions") },
111 { "new", OPT_NEW, NULL, 0,
112 N_("Do not use existing catalog, force new output file") },
113 { "output", 'o', N_("NAME"), 0, N_("Write output to file NAME") },
114 { NULL, 0, NULL, 0, NULL }
115};
116
117/* Short description of program. */
118static const char doc[] = N_("Generate message catalog.\
119\vIf INPUT-FILE is -, input is read from standard input. If OUTPUT-FILE\n\
120is -, output is written to standard output.\n");
121
122/* Strings for arguments in help texts. */
123static const char args_doc[] = N_("\
124-o OUTPUT-FILE [INPUT-FILE]...\n[OUTPUT-FILE [INPUT-FILE]...]");
125
126/* Prototype for option handler. */
127static error_t parse_opt (int key, char *arg, struct argp_state *state);
128
129/* Function to print some extra text in the help message. */
130static char *more_help (int key, const char *text, void *input);
131
132/* Data structure to communicate with argp functions. */
133static struct argp argp =
134{
135 options, parse_opt, args_doc, doc, NULL, more_help
136};
137
138
139/* Wrapper functions with error checking for standard functions. */
140#include <programs/xmalloc.h>
141
142/* Prototypes for local functions. */
143static void error_print (void);
144static struct catalog *read_input_file (struct catalog *current,
145 const char *fname);
146static void write_out (struct catalog *result, const char *output_name,
147 const char *header_name);
148static struct set_list *find_set (struct catalog *current, int number);
149static void normalize_line (const char *fname, size_t line, iconv_t cd,
150 wchar_t *string, wchar_t quote_char,
151 wchar_t escape_char);
152static void read_old (struct catalog *catalog, const char *file_name);
153static int open_conversion (const char *codesetp, iconv_t *cd_towcp,
154 iconv_t *cd_tombp, wchar_t *escape_charp);
155
156
157int
158main (int argc, char *argv[])
159{
160 struct catalog *result;
161 int remaining;
162
163 /* Set program name for messages. */
164 error_print_progname = error_print;
165
166 /* Set locale via LC_ALL. */
167 setlocale (LC_ALL, "");
168
169 /* Set the text message domain. */
170 textdomain (PACKAGE);
171
172 /* Initialize local variables. */
173 result = NULL;
174
175 /* Parse and process arguments. */
176 argp_parse (&argp, argc, argv, 0, &remaining, NULL);
177
178 /* Determine output file. */
179 if (output_name == NULL)
180 output_name = remaining < argc ? argv[remaining++] : "-";
181
182 /* Process all input files. */
183 setlocale (LC_CTYPE, "C");
184 if (remaining < argc)
185 do
186 result = read_input_file (result, argv[remaining]);
187 while (++remaining < argc);
188 else
189 result = read_input_file (NULL, "-");
190
191 /* Write out the result. */
192 if (result != NULL)
193 write_out (result, output_name, header_name);
194
195 return error_message_count != 0;
196}
197
198
199/* Handle program arguments. */
200static error_t
201parse_opt (int key, char *arg, struct argp_state *state)
202{
203 switch (key)
204 {
205 case 'H':
206 header_name = arg;
207 break;
208 case OPT_NEW:
209 force_new = 1;
210 break;
211 case 'o':
212 output_name = arg;
213 break;
214 default:
215 return ARGP_ERR_UNKNOWN;
216 }
217 return 0;
218}
219
220
221static char *
222more_help (int key, const char *text, void *input)
223{
224 char *tp = NULL;
225 switch (key)
226 {
227 case ARGP_KEY_HELP_EXTRA:
228 /* We print some extra information. */
229 if (asprintf (&tp, gettext ("\
230For bug reporting instructions, please see:\n\
231%s.\n"), REPORT_BUGS_TO) < 0)
232 return NULL;
233 return tp;
234 default:
235 break;
236 }
237 return (char *) text;
238}
239
240/* Print the version information. */
241static void
242print_version (FILE *stream, struct argp_state *state)
243{
244 fprintf (stream, "gencat %s%s\n", PKGVERSION, VERSION);
245 fprintf (stream, gettext ("\
246Copyright (C) %s Free Software Foundation, Inc.\n\
247This is free software; see the source for copying conditions. There is NO\n\
248warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n\
249"), "2023");
250 fprintf (stream, gettext ("Written by %s.\n"), "Ulrich Drepper");
251}
252
253
254/* The address of this function will be assigned to the hook in the
255 error functions. */
256static void
257error_print (void)
258{
259 /* We don't want the program name to be printed in messages. Emacs'
260 compile.el does not like this. */
261}
262
263
264static struct catalog *
265read_input_file (struct catalog *current, const char *fname)
266{
267 FILE *fp;
268 char *buf;
269 size_t len;
270 size_t line_number;
271 wchar_t *wbuf;
272 size_t wbufsize;
273 iconv_t cd_towc = (iconv_t) -1;
274 iconv_t cd_tomb = (iconv_t) -1;
275 wchar_t escape_char = L'\\';
276 char *codeset = NULL;
277
278 if (strcmp (fname, "-") == 0 || strcmp (fname, "/dev/stdin") == 0)
279 {
280 fp = stdin;
281 fname = gettext ("*standard input*");
282 }
283 else
284 fp = fopen (fname, "r");
285 if (fp == NULL)
286 {
287 error (0, errno, gettext ("cannot open input file `%s'"), fname);
288 return current;
289 }
290
291 /* If we haven't seen anything yet, allocate result structure. */
292 if (current == NULL)
293 {
294 current = (struct catalog *) xcalloc (1, sizeof (*current));
295
296#define obstack_chunk_alloc malloc
297#define obstack_chunk_free free
298 obstack_init (&current->mem_pool);
299
300 current->current_set = find_set (current, NL_SETD);
301 }
302
303 buf = NULL;
304 len = 0;
305 line_number = 0;
306
307 wbufsize = 1024;
308 wbuf = (wchar_t *) xmalloc (wbufsize);
309
310 while (!feof (fp))
311 {
312 int continued;
313 int used;
314 size_t start_line = line_number + 1;
315 char *this_line;
316
317 do
318 {
319 int act_len;
320
321 act_len = getline (&buf, &len, fp);
322 if (act_len <= 0)
323 break;
324 ++line_number;
325
326 /* It the line continued? */
327 continued = 0;
328 if (buf[act_len - 1] == '\n')
329 {
330 --act_len;
331
332 /* There might be more than one backslash at the end of
333 the line. Only if there is an odd number of them is
334 the line continued. */
335 if (act_len > 0 && buf[act_len - 1] == '\\')
336 {
337 int temp_act_len = act_len;
338
339 do
340 {
341 --temp_act_len;
342 continued = !continued;
343 }
344 while (temp_act_len > 0 && buf[temp_act_len - 1] == '\\');
345
346 if (continued)
347 --act_len;
348 }
349 }
350
351 /* Append to currently selected line. */
352 obstack_grow (&current->mem_pool, buf, act_len);
353 }
354 while (continued);
355
356 obstack_1grow (&current->mem_pool, '\0');
357 this_line = (char *) obstack_finish (&current->mem_pool);
358
359 used = 0;
360 if (this_line[0] == '$')
361 {
362 if (isblank (this_line[1]))
363 {
364 int cnt = 1;
365 while (isblank (this_line[cnt]))
366 ++cnt;
367 if (strncmp (&this_line[cnt], "codeset=", 8) != 0)
368 /* This is a comment line. Do nothing. */;
369 else if (codeset != NULL)
370 /* Ignore multiple codeset. */;
371 else
372 {
373 int start = cnt + 8;
374 cnt = start;
375 while (this_line[cnt] != '\0' && !isspace (this_line[cnt]))
376 ++cnt;
377 if (cnt != start)
378 {
379 int len = cnt - start;
380 codeset = xmalloc (len + 1);
381 *((char *) mempcpy (codeset, &this_line[start], len))
382 = '\0';
383 }
384 }
385 }
386 else if (strncmp (&this_line[1], "set", 3) == 0)
387 {
388 int cnt = sizeof ("set");
389 int set_number;
390 const char *symbol = NULL;
391 while (isspace (this_line[cnt]))
392 ++cnt;
393
394 if (isdigit (this_line[cnt]))
395 {
396 set_number = atol (&this_line[cnt]);
397
398 /* If the given number for the character set is
399 higher than any we used for symbolic set names
400 avoid clashing by using only higher numbers for
401 the following symbolic definitions. */
402 if (set_number > current->last_set)
403 current->last_set = set_number;
404 }
405 else
406 {
407 /* See whether it is a reasonable identifier. */
408 int start = cnt;
409 while (isalnum (this_line[cnt]) || this_line[cnt] == '_')
410 ++cnt;
411
412 if (cnt == start)
413 {
414 /* No correct character found. */
415 error_at_line (0, 0, fname, start_line,
416 gettext ("illegal set number"));
417 set_number = 0;
418 }
419 else
420 {
421 /* We have found seomthing that looks like a
422 correct identifier. */
423 struct set_list *runp;
424
425 this_line[cnt] = '\0';
426 used = 1;
427 symbol = &this_line[start];
428
429 /* Test whether the identifier was already used. */
430 runp = current->all_sets;
431 while (runp != 0)
432 if (runp->symbol != NULL
433 && strcmp (runp->symbol, symbol) == 0)
434 break;
435 else
436 runp = runp->next;
437
438 if (runp != NULL)
439 {
440 /* We cannot allow duplicate identifiers for
441 message sets. */
442 error_at_line (0, 0, fname, start_line,
443 gettext ("duplicate set definition"));
444 error_at_line (0, 0, runp->fname, runp->line,
445 gettext ("\
446this is the first definition"));
447 set_number = 0;
448 }
449 else
450 /* Allocate next free message set for identifier. */
451 set_number = ++current->last_set;
452 }
453 }
454
455 if (set_number != 0)
456 {
457 /* We found a legal set number. */
458 current->current_set = find_set (current, set_number);
459 if (symbol != NULL)
460 used = 1;
461 current->current_set->symbol = symbol;
462 current->current_set->fname = fname;
463 current->current_set->line = start_line;
464 }
465 }
466 else if (strncmp (&this_line[1], "delset", 6) == 0)
467 {
468 int cnt = sizeof ("delset");
469 while (isspace (this_line[cnt]))
470 ++cnt;
471
472 if (isdigit (this_line[cnt]))
473 {
474 size_t set_number = atol (&this_line[cnt]);
475 struct set_list *set;
476
477 /* Mark the message set with the given number as
478 deleted. */
479 set = find_set (current, set_number);
480 set->deleted = 1;
481 }
482 else
483 {
484 /* See whether it is a reasonable identifier. */
485 int start = cnt;
486 while (isalnum (this_line[cnt]) || this_line[cnt] == '_')
487 ++cnt;
488
489 if (cnt == start)
490 error_at_line (0, 0, fname, start_line,
491 gettext ("illegal set number"));
492 else
493 {
494 const char *symbol;
495 struct set_list *runp;
496
497 this_line[cnt] = '\0';
498 used = 1;
499 symbol = &this_line[start];
500
501 /* We have a symbolic set name. This name must
502 appear somewhere else in the catalogs read so
503 far. */
504 for (runp = current->all_sets; runp != NULL;
505 runp = runp->next)
506 {
507 if (strcmp (runp->symbol, symbol) == 0)
508 {
509 runp->deleted = 1;
510 break;
511 }
512 }
513 if (runp == NULL)
514 /* Name does not exist before. */
515 error_at_line (0, 0, fname, start_line,
516 gettext ("unknown set `%s'"), symbol);
517 }
518 }
519 }
520 else if (strncmp (&this_line[1], "quote", 5) == 0)
521 {
522 char buf[2];
523 char *bufptr;
524 size_t buflen;
525 char *wbufptr;
526 size_t wbuflen;
527 int cnt;
528
529 cnt = sizeof ("quote");
530 while (isspace (this_line[cnt]))
531 ++cnt;
532
533 /* We need the conversion. */
534 if (cd_towc == (iconv_t) -1
535 && open_conversion (codeset, &cd_towc, &cd_tomb,
536 &escape_char) != 0)
537 /* Something is wrong. */
538 goto out;
539
540 /* Yes, the quote char can be '\0'; this means no quote
541 char. The function using the information works on
542 wide characters so we have to convert it here. */
543 buf[0] = this_line[cnt];
544 buf[1] = '\0';
545 bufptr = buf;
546 buflen = 2;
547
548 wbufptr = (char *) wbuf;
549 wbuflen = wbufsize;
550
551 /* Flush the state. */
552 iconv (cd_towc, NULL, NULL, NULL, NULL);
553
554 iconv (cd_towc, &bufptr, &buflen, &wbufptr, &wbuflen);
555 if (buflen != 0 || (wchar_t *) wbufptr != &wbuf[2])
556 error_at_line (0, 0, fname, start_line,
557 gettext ("invalid quote character"));
558 else
559 /* Use the converted wide character. */
560 current->quote_char = wbuf[0];
561 }
562 else
563 {
564 int cnt;
565 cnt = 2;
566 while (this_line[cnt] != '\0' && !isspace (this_line[cnt]))
567 ++cnt;
568 this_line[cnt] = '\0';
569 error_at_line (0, 0, fname, start_line,
570 gettext ("unknown directive `%s': line ignored"),
571 &this_line[1]);
572 }
573 }
574 else if (isalnum (this_line[0]) || this_line[0] == '_')
575 {
576 const char *ident = this_line;
577 char *line = this_line;
578 int message_number;
579
580 do
581 ++line;
582 while (line[0] != '\0' && !isspace (line[0]));
583 if (line[0] != '\0')
584 *line++ = '\0'; /* Terminate the identifier. */
585
586 /* Now we found the beginning of the message itself. */
587
588 if (isdigit (ident[0]))
589 {
590 struct message_list *runp;
591 struct message_list *lastp;
592
593 message_number = atoi (ident);
594
595 /* Find location to insert the new message. */
596 runp = current->current_set->messages;
597 lastp = NULL;
598 while (runp != NULL)
599 if (runp->number == message_number)
600 break;
601 else
602 {
603 lastp = runp;
604 runp = runp->next;
605 }
606 if (runp != NULL)
607 {
608 /* Oh, oh. There is already a message with this
609 number in the message set. */
610 if (runp->symbol == NULL)
611 {
612 /* The existing message had its number specified
613 by the user. Fatal collision type uh, oh. */
614 error_at_line (0, 0, fname, start_line,
615 gettext ("duplicated message number"));
616 error_at_line (0, 0, runp->fname, runp->line,
617 gettext ("this is the first definition"));
618 message_number = 0;
619 }
620 else
621 {
622 /* Collision was with number auto-assigned to a
623 symbolic. Change existing symbolic number
624 and move to end the list (if not already there). */
625 runp->number = ++current->current_set->last_message;
626
627 if (runp->next != NULL)
628 {
629 struct message_list *endp;
630
631 if (lastp == NULL)
632 current->current_set->messages=runp->next;
633 else
634 lastp->next=runp->next;
635
636 endp = runp->next;
637 while (endp->next != NULL)
638 endp = endp->next;
639
640 endp->next = runp;
641 runp->next = NULL;
642 }
643 }
644 }
645 ident = NULL; /* We don't have a symbol. */
646
647 if (message_number != 0
648 && message_number > current->current_set->last_message)
649 current->current_set->last_message = message_number;
650 }
651 else if (ident[0] != '\0')
652 {
653 struct message_list *runp;
654
655 /* Test whether the symbolic name was not used for
656 another message in this message set. */
657 runp = current->current_set->messages;
658 while (runp != NULL)
659 if (runp->symbol != NULL && strcmp (ident, runp->symbol) == 0)
660 break;
661 else
662 runp = runp->next;
663 if (runp != NULL)
664 {
665 /* The name is already used. */
666 error_at_line (0, 0, fname, start_line, gettext ("\
667duplicated message identifier"));
668 error_at_line (0, 0, runp->fname, runp->line,
669 gettext ("this is the first definition"));
670 message_number = 0;
671 }
672 else
673 /* Give the message the next unused number. */
674 message_number = ++current->current_set->last_message;
675 }
676 else
677 message_number = 0;
678
679 if (message_number != 0)
680 {
681 char *inbuf;
682 size_t inlen;
683 char *outbuf;
684 size_t outlen;
685 struct message_list *newp;
686 size_t line_len = strlen (line) + 1;
687 size_t ident_len = 0;
688
689 /* We need the conversion. */
690 if (cd_towc == (iconv_t) -1
691 && open_conversion (codeset, &cd_towc, &cd_tomb,
692 &escape_char) != 0)
693 /* Something is wrong. */
694 goto out;
695
696 /* Convert to a wide character string. We have to
697 interpret escape sequences which will be impossible
698 without doing the conversion if the codeset of the
699 message is stateful. */
700 while (1)
701 {
702 inbuf = line;
703 inlen = line_len;
704 outbuf = (char *) wbuf;
705 outlen = wbufsize;
706
707 /* Flush the state. */
708 iconv (cd_towc, NULL, NULL, NULL, NULL);
709
710 iconv (cd_towc, &inbuf, &inlen, &outbuf, &outlen);
711 if (inlen == 0)
712 {
713 /* The string is converted. */
714 assert (outlen < wbufsize);
715 assert (wbuf[(wbufsize - outlen) / sizeof (wchar_t) - 1]
716 == L'\0');
717 break;
718 }
719
720 if (outlen != 0)
721 {
722 /* Something is wrong with this string, we ignore it. */
723 error_at_line (0, 0, fname, start_line, gettext ("\
724invalid character: message ignored"));
725 goto ignore;
726 }
727
728 /* The output buffer is too small. */
729 wbufsize *= 2;
730 wbuf = (wchar_t *) xrealloc (wbuf, wbufsize);
731 }
732
733 /* Strip quote characters, change escape sequences into
734 correct characters etc. */
735 normalize_line (fname, start_line, cd_towc, wbuf,
736 current->quote_char, escape_char);
737
738 if (ident)
739 ident_len = line - this_line;
740
741 /* Now the string is free of escape sequences. Convert it
742 back into a multibyte character string. First free the
743 memory allocated for the original string. */
744 obstack_free (&current->mem_pool, this_line);
745
746 used = 1; /* Yes, we use the line. */
747
748 /* Now fill in the new string. It should never happen that
749 the replaced string is longer than the original. */
750 inbuf = (char *) wbuf;
751 inlen = (wcslen (wbuf) + 1) * sizeof (wchar_t);
752
753 outlen = obstack_room (&current->mem_pool);
754 obstack_blank (&current->mem_pool, outlen);
755 this_line = (char *) obstack_base (&current->mem_pool);
756 outbuf = this_line + ident_len;
757 outlen -= ident_len;
758
759 /* Flush the state. */
760 iconv (cd_tomb, NULL, NULL, NULL, NULL);
761
762 iconv (cd_tomb, &inbuf, &inlen, &outbuf, &outlen);
763 if (inlen != 0)
764 {
765 error_at_line (0, 0, fname, start_line,
766 gettext ("invalid line"));
767 goto ignore;
768 }
769 assert (outbuf[-1] == '\0');
770
771 /* Free the memory in the obstack we don't use. */
772 obstack_blank (&current->mem_pool, -(int) outlen);
773 line = obstack_finish (&current->mem_pool);
774
775 newp = (struct message_list *) xmalloc (sizeof (*newp));
776 newp->number = message_number;
777 newp->message = line + ident_len;
778 /* Remember symbolic name; is NULL if no is given. */
779 newp->symbol = ident ? line : NULL;
780 /* Remember where we found the character. */
781 newp->fname = fname;
782 newp->line = start_line;
783
784 /* Find place to insert to message. We keep them in a
785 sorted single linked list. */
786 if (current->current_set->messages == NULL
787 || current->current_set->messages->number > message_number)
788 {
789 newp->next = current->current_set->messages;
790 current->current_set->messages = newp;
791 }
792 else
793 {
794 struct message_list *runp;
795 runp = current->current_set->messages;
796 while (runp->next != NULL)
797 if (runp->next->number > message_number)
798 break;
799 else
800 runp = runp->next;
801 newp->next = runp->next;
802 runp->next = newp;
803 }
804 }
805 ++current->total_messages;
806 }
807 else
808 {
809 size_t cnt;
810
811 cnt = 0;
812 /* See whether we have any non-white space character in this
813 line. */
814 while (this_line[cnt] != '\0' && isspace (this_line[cnt]))
815 ++cnt;
816
817 if (this_line[cnt] != '\0')
818 /* Yes, some unknown characters found. */
819 error_at_line (0, 0, fname, start_line,
820 gettext ("malformed line ignored"));
821 }
822
823 ignore:
824 /* We can save the memory for the line if it was not used. */
825 if (!used)
826 obstack_free (&current->mem_pool, this_line);
827 }
828
829 /* Close the conversion modules. */
830 iconv_close (cd_towc);
831 iconv_close (cd_tomb);
832 free (codeset);
833
834 out:
835 free (wbuf);
836
837 if (fp != stdin)
838 fclose (fp);
839 return current;
840}
841
842static void
843write_out (struct catalog *catalog, const char *output_name,
844 const char *header_name)
845{
846 /* Computing the "optimal" size. */
847 struct set_list *set_run;
848 size_t best_total, best_size, best_depth;
849 size_t act_size, act_depth;
850 struct catalog_obj obj;
851 struct obstack string_pool;
852 const char *strings;
853 size_t strings_size;
854 uint32_t *array1, *array2;
855 size_t cnt;
856 int fd;
857
858 /* If not otherwise told try to read file with existing
859 translations. */
860 if (!force_new)
861 read_old (catalog, output_name);
862
863 /* Initialize best_size with a very high value. */
864 best_total = best_size = best_depth = UINT_MAX;
865
866 /* We need some start size for testing. Let's start with
867 TOTAL_MESSAGES / 5, which theoretically provides a mean depth of
868 5. */
869 act_size = 1 + catalog->total_messages / 5;
870
871 /* We determine the size of a hash table here. Because the message
872 numbers can be chosen arbitrary by the programmer we cannot use
873 the simple method of accessing the array using the message
874 number. The algorithm is based on the trivial hash function
875 NUMBER % TABLE_SIZE, where collisions are stored in a second
876 dimension up to TABLE_DEPTH. We here compute TABLE_SIZE so that
877 the needed space (= TABLE_SIZE * TABLE_DEPTH) is minimal. */
878 while (act_size <= best_total)
879 {
880 size_t deep[act_size];
881
882 act_depth = 1;
883 memset (deep, '\0', act_size * sizeof (size_t));
884 set_run = catalog->all_sets;
885 while (set_run != NULL)
886 {
887 struct message_list *message_run;
888
889 message_run = set_run->messages;
890 while (message_run != NULL)
891 {
892 size_t idx = (message_run->number * set_run->number) % act_size;
893
894 ++deep[idx];
895 if (deep[idx] > act_depth)
896 {
897 act_depth = deep[idx];
898 if (act_depth * act_size > best_total)
899 break;
900 }
901 message_run = message_run->next;
902 }
903 set_run = set_run->next;
904 }
905
906 if (act_depth * act_size <= best_total)
907 {
908 /* We have found a better solution. */
909 best_total = act_depth * act_size;
910 best_size = act_size;
911 best_depth = act_depth;
912 }
913
914 ++act_size;
915 }
916
917 /* let's be prepared for an empty message file. */
918 if (best_size == UINT_MAX)
919 {
920 best_size = 1;
921 best_depth = 1;
922 }
923
924 /* OK, now we have the size we will use. Fill in the header, build
925 the table and the second one with swapped byte order. */
926 obj.magic = CATGETS_MAGIC;
927 obj.plane_size = best_size;
928 obj.plane_depth = best_depth;
929
930 uint32_t array_size = best_size * best_depth * sizeof (uint32_t) * 3;
931 /* Allocate room for all needed arrays. */
932 array1 = (uint32_t *) alloca (array_size);
933 memset (array1, '\0', array_size);
934 array2 = (uint32_t *) alloca (array_size);
935 obstack_init (&string_pool);
936
937 set_run = catalog->all_sets;
938 while (set_run != NULL)
939 {
940 struct message_list *message_run;
941
942 message_run = set_run->messages;
943 while (message_run != NULL)
944 {
945 size_t idx = (((message_run->number * set_run->number) % best_size)
946 * 3);
947 /* Determine collision depth. */
948 while (array1[idx] != 0)
949 idx += best_size * 3;
950
951 /* Store set number, message number and pointer into string
952 space, relative to the first string. */
953 array1[idx + 0] = set_run->number;
954 array1[idx + 1] = message_run->number;
955 array1[idx + 2] = obstack_object_size (&string_pool);
956
957 /* Add current string to the continuous space containing all
958 strings. */
959 obstack_grow0 (&string_pool, message_run->message,
960 strlen (message_run->message));
961
962 message_run = message_run->next;
963 }
964
965 set_run = set_run->next;
966 }
967 strings_size = obstack_object_size (&string_pool);
968 strings = obstack_finish (&string_pool);
969
970 /* Compute ARRAY2 by changing the byte order. */
971 for (cnt = 0; cnt < best_size * best_depth * 3; ++cnt)
972 array2[cnt] = SWAPU32 (array1[cnt]);
973
974 /* Now we can write out the whole data. */
975 if (strcmp (output_name, "-") == 0
976 || strcmp (output_name, "/dev/stdout") == 0)
977 fd = STDOUT_FILENO;
978 else
979 {
980 fd = creat (output_name, 0666);
981 if (fd < 0)
982 error (EXIT_FAILURE, errno, gettext ("cannot open output file `%s'"),
983 output_name);
984 }
985
986 /* Write out header. */
987 write_all(fd, &obj, sizeof (obj));
988
989 /* We always write out the little endian version of the index
990 arrays. */
991#if __BYTE_ORDER == __LITTLE_ENDIAN
992 write_all(fd, array1, array_size);
993 write_all(fd, array2, array_size);
994#elif __BYTE_ORDER == __BIG_ENDIAN
995 write_all(fd, array2, array_size);
996 write_all(fd, array1, array_size);
997#else
998# error Cannot handle __BYTE_ORDER byte order
999#endif
1000
1001 /* Finally write the strings. */
1002 write_all(fd, strings, strings_size);
1003
1004 if (fd != STDOUT_FILENO)
1005 close (fd);
1006
1007 /* If requested now write out the header file. */
1008 if (header_name != NULL)
1009 {
1010 int first = 1;
1011 FILE *fp;
1012
1013 /* Open output file. "-" or "/dev/stdout" means write to
1014 standard output. */
1015 if (strcmp (header_name, "-") == 0
1016 || strcmp (header_name, "/dev/stdout") == 0)
1017 fp = stdout;
1018 else
1019 {
1020 fp = fopen (header_name, "w");
1021 if (fp == NULL)
1022 error (EXIT_FAILURE, errno,
1023 gettext ("cannot open output file `%s'"), header_name);
1024 }
1025
1026 /* Iterate over all sets and all messages. */
1027 set_run = catalog->all_sets;
1028 while (set_run != NULL)
1029 {
1030 struct message_list *message_run;
1031
1032 /* If the current message set has a symbolic name write this
1033 out first. */
1034 if (set_run->symbol != NULL)
1035 fprintf (fp, "%s#define %sSet %#x\t/* %s:%zu */\n",
1036 first ? "" : "\n", set_run->symbol, set_run->number - 1,
1037 set_run->fname, set_run->line);
1038 first = 0;
1039
1040 message_run = set_run->messages;
1041 while (message_run != NULL)
1042 {
1043 /* If the current message has a symbolic name write
1044 #define out. But we have to take care for the set
1045 not having a symbolic name. */
1046 if (message_run->symbol != NULL)
1047 {
1048 if (set_run->symbol == NULL)
1049 fprintf (fp, "#define AutomaticSet%d%s %#x\t/* %s:%zu */\n",
1050 set_run->number, message_run->symbol,
1051 message_run->number, message_run->fname,
1052 message_run->line);
1053 else
1054 fprintf (fp, "#define %s%s %#x\t/* %s:%zu */\n",
1055 set_run->symbol, message_run->symbol,
1056 message_run->number, message_run->fname,
1057 message_run->line);
1058 }
1059
1060 message_run = message_run->next;
1061 }
1062
1063 set_run = set_run->next;
1064 }
1065
1066 if (fp != stdout)
1067 fclose (fp);
1068 }
1069}
1070
1071
1072static struct set_list *
1073find_set (struct catalog *current, int number)
1074{
1075 struct set_list *result = current->all_sets;
1076
1077 /* We must avoid set number 0 because a set of this number signals
1078 in the tables that the entry is not occupied. */
1079 ++number;
1080
1081 while (result != NULL)
1082 if (result->number == number)
1083 return result;
1084 else
1085 result = result->next;
1086
1087 /* Prepare new message set. */
1088 result = (struct set_list *) xcalloc (1, sizeof (*result));
1089 result->number = number;
1090 result->next = current->all_sets;
1091 current->all_sets = result;
1092
1093 return result;
1094}
1095
1096
1097/* Normalize given string *in*place* by processing escape sequences
1098 and quote characters. */
1099static void
1100normalize_line (const char *fname, size_t line, iconv_t cd, wchar_t *string,
1101 wchar_t quote_char, wchar_t escape_char)
1102{
1103 int is_quoted;
1104 wchar_t *rp = string;
1105 wchar_t *wp = string;
1106
1107 if (quote_char != L'\0' && *rp == quote_char)
1108 {
1109 is_quoted = 1;
1110 ++rp;
1111 }
1112 else
1113 is_quoted = 0;
1114
1115 while (*rp != L'\0')
1116 if (*rp == quote_char)
1117 /* We simply end the string when we find the first time an
1118 not-escaped quote character. */
1119 break;
1120 else if (*rp == escape_char)
1121 {
1122 ++rp;
1123 if (quote_char != L'\0' && *rp == quote_char)
1124 /* This is an extension to XPG. */
1125 *wp++ = *rp++;
1126 else
1127 /* Recognize escape sequences. */
1128 switch (*rp)
1129 {
1130 case L'n':
1131 *wp++ = L'\n';
1132 ++rp;
1133 break;
1134 case L't':
1135 *wp++ = L'\t';
1136 ++rp;
1137 break;
1138 case L'v':
1139 *wp++ = L'\v';
1140 ++rp;
1141 break;
1142 case L'b':
1143 *wp++ = L'\b';
1144 ++rp;
1145 break;
1146 case L'r':
1147 *wp++ = L'\r';
1148 ++rp;
1149 break;
1150 case L'f':
1151 *wp++ = L'\f';
1152 ++rp;
1153 break;
1154 case L'0' ... L'7':
1155 {
1156 int number;
1157 char cbuf[2];
1158 char *cbufptr;
1159 size_t cbufin;
1160 wchar_t wcbuf[2];
1161 char *wcbufptr;
1162 size_t wcbufin;
1163
1164 number = *rp++ - L'0';
1165 while (number <= (255 / 8) && *rp >= L'0' && *rp <= L'7')
1166 {
1167 number *= 8;
1168 number += *rp++ - L'0';
1169 }
1170
1171 cbuf[0] = (char) number;
1172 cbuf[1] = '\0';
1173 cbufptr = cbuf;
1174 cbufin = 2;
1175
1176 wcbufptr = (char *) wcbuf;
1177 wcbufin = sizeof (wcbuf);
1178
1179 /* Flush the state. */
1180 iconv (cd, NULL, NULL, NULL, NULL);
1181
1182 iconv (cd, &cbufptr, &cbufin, &wcbufptr, &wcbufin);
1183 if (cbufptr != &cbuf[2] || (wchar_t *) wcbufptr != &wcbuf[2])
1184 error_at_line (0, 0, fname, line,
1185 gettext ("invalid escape sequence"));
1186 else
1187 *wp++ = wcbuf[0];
1188 }
1189 break;
1190 default:
1191 if (*rp == escape_char)
1192 {
1193 *wp++ = escape_char;
1194 ++rp;
1195 }
1196 else
1197 {
1198 /* Simply ignore the backslash character. */
1199 }
1200 break;
1201 }
1202 }
1203 else
1204 *wp++ = *rp++;
1205
1206 /* If we saw a quote character at the beginning we expect another
1207 one at the end. */
1208 if (is_quoted && *rp != quote_char)
1209 error_at_line (0, 0, fname, line, gettext ("unterminated message"));
1210
1211 /* Terminate string. */
1212 *wp = L'\0';
1213 return;
1214}
1215
1216
1217static void
1218read_old (struct catalog *catalog, const char *file_name)
1219{
1220 struct catalog_info old_cat_obj;
1221 struct set_list *set = NULL;
1222 int last_set = -1;
1223 size_t cnt;
1224
1225 /* Try to open catalog, but don't look through the NLSPATH. */
1226 if (__open_catalog (file_name, NULL, NULL, &old_cat_obj) != 0)
1227 {
1228 if (errno == ENOENT)
1229 /* No problem, the catalog simply does not exist. */
1230 return;
1231 else
1232 error (EXIT_FAILURE, errno,
1233 gettext ("while opening old catalog file"));
1234 }
1235
1236 /* OK, we have the catalog loaded. Now read all messages and merge
1237 them. When set and message number clash for any message the new
1238 one is used. If the new one is empty it indicates that the
1239 message should be deleted. */
1240 for (cnt = 0; cnt < old_cat_obj.plane_size * old_cat_obj.plane_depth; ++cnt)
1241 {
1242 struct message_list *message, *last;
1243
1244 if (old_cat_obj.name_ptr[cnt * 3 + 0] == 0)
1245 /* No message in this slot. */
1246 continue;
1247
1248 if (old_cat_obj.name_ptr[cnt * 3 + 0] - 1 != (uint32_t) last_set)
1249 {
1250 last_set = old_cat_obj.name_ptr[cnt * 3 + 0] - 1;
1251 set = find_set (catalog, old_cat_obj.name_ptr[cnt * 3 + 0] - 1);
1252 }
1253
1254 last = NULL;
1255 message = set->messages;
1256 while (message != NULL)
1257 {
1258 if ((uint32_t) message->number >= old_cat_obj.name_ptr[cnt * 3 + 1])
1259 break;
1260 last = message;
1261 message = message->next;
1262 }
1263
1264 if (message == NULL
1265 || (uint32_t) message->number > old_cat_obj.name_ptr[cnt * 3 + 1])
1266 {
1267 /* We have found a message which is not yet in the catalog.
1268 Insert it at the right position. */
1269 struct message_list *newp;
1270
1271 newp = (struct message_list *) xmalloc (sizeof (*newp));
1272 newp->number = old_cat_obj.name_ptr[cnt * 3 + 1];
1273 newp->message =
1274 &old_cat_obj.strings[old_cat_obj.name_ptr[cnt * 3 + 2]];
1275 newp->fname = NULL;
1276 newp->line = 0;
1277 newp->symbol = NULL;
1278 newp->next = message;
1279
1280 if (last == NULL)
1281 set->messages = newp;
1282 else
1283 last->next = newp;
1284
1285 ++catalog->total_messages;
1286 }
1287 else if (*message->message == '\0')
1288 {
1289 /* The new empty message has overridden the old one thus
1290 "deleting" it as required. Now remove the empty remains. */
1291 if (last == NULL)
1292 set->messages = message->next;
1293 else
1294 last->next = message->next;
1295 }
1296 }
1297}
1298
1299
1300static int
1301open_conversion (const char *codeset, iconv_t *cd_towcp, iconv_t *cd_tombp,
1302 wchar_t *escape_charp)
1303{
1304 char buf[2];
1305 char *bufptr;
1306 size_t bufsize;
1307 wchar_t wbuf[2];
1308 char *wbufptr;
1309 size_t wbufsize;
1310
1311 /* If the input file does not specify the codeset use the locale's. */
1312 if (codeset == NULL)
1313 {
1314 setlocale (LC_ALL, "");
1315 codeset = nl_langinfo (CODESET);
1316 setlocale (LC_ALL, "C");
1317 }
1318
1319 /* Get the conversion modules. */
1320 *cd_towcp = iconv_open ("WCHAR_T", codeset);
1321 *cd_tombp = iconv_open (codeset, "WCHAR_T");
1322 if (*cd_towcp == (iconv_t) -1 || *cd_tombp == (iconv_t) -1)
1323 {
1324 error (0, 0, gettext ("conversion modules not available"));
1325 if (*cd_towcp != (iconv_t) -1)
1326 iconv_close (*cd_towcp);
1327
1328 return 1;
1329 }
1330
1331 /* One special case for historical reasons is the backslash
1332 character. In some codesets the byte value 0x5c is not mapped to
1333 U005c in Unicode. These charsets then don't have a backslash
1334 character at all. Therefore we have to live with whatever the
1335 codeset provides and recognize, instead of the U005c, the character
1336 the byte value 0x5c is mapped to. */
1337 buf[0] = '\\';
1338 buf[1] = '\0';
1339 bufptr = buf;
1340 bufsize = 2;
1341
1342 wbufptr = (char *) wbuf;
1343 wbufsize = sizeof (wbuf);
1344
1345 iconv (*cd_towcp, &bufptr, &bufsize, &wbufptr, &wbufsize);
1346 if (bufsize != 0 || wbufsize != 0)
1347 {
1348 /* Something went wrong, we couldn't convert the byte 0x5c. Go
1349 on with using U005c. */
1350 error (0, 0, gettext ("cannot determine escape character"));
1351 *escape_charp = L'\\';
1352 }
1353 else
1354 *escape_charp = wbuf[0];
1355
1356 return 0;
1357}
1358